AI Travel Stack: Five Competing Layers, One Platform-Level Threat

The major OTAs have each placed a different bet on which layer of the AI stack is worth owning. The risk is that all of them lose to Apple and Google.

Context

AI is forcing a structural realignment in travel distribution. The question is no longer whether OTAs will deploy AI — they all are. The question is which layer of the AI stack determines the customer relationship as AI becomes the primary interface for search, booking, and service.

Skift’s analysis of the current competitive landscape identifies five layers where value may concentrate: model, orchestration, product, legibility, and OS/platform. The major players in travel have organized their AI strategies around these layers — and the divergence is now concrete, not theoretical.

The Detail

Model layer: Booking Holdings and Trip.com are building proprietary LLMs trained on their own transaction data. Booking’s theory is that travel-domain intelligence — pricing signals, booking patterns, consumer intent at scale — creates a defensible edge that general-purpose foundation models can’t replicate. Trip.com is pursuing an analogous strategy in Asian markets. Both bets assume that owning the model means owning the intelligence advantage.

Orchestration and product layers: Expedia has positioned itself as an orchestration-layer player — building the workflow infrastructure that routes AI-generated queries into bookable inventory across suppliers. The bet: the plumbing layer, not the model, becomes the sticky differentiation point as agents proliferate. Airbnb targets the product layer instead, investing in making the actual experience surface — search, stays, Experiences — the AI-native interface that keeps customers in the Airbnb environment rather than routing through an intermediary.

Legibility layer: Sabre occupies the infrastructure position. Airlines and travel agencies have decades of inventory structured in GDS formats that current AI systems struggle to parse natively. Sabre’s legibility-layer work addresses the translation problem: converting between AI-generated intent and structured inventory systems that weren’t built for natural language interfaces. This is unglamorous work, but the value chain above it breaks if the translation fails.

The threat above all five layers is the OS/platform layer. Apple Intelligence and Google Gemini are already embedded in the device and search interfaces that travelers use before reaching any OTA. As agentic AI matures in travel — AI that books rather than just searches — the entity controlling the task-initiation interface controls the customer relationship. That entity is increasingly Apple or Google, not an OTA.

Source: Skift

Content created with AI assistance and reviewed for accuracy.